Problem
As the team grew, code review throughput became a bottleneck. Senior engineers spent their best hours on the predictable parts (naming, test coverage, config drift) instead of architecture.
Architecture
flowchart LR MR[GitLab MR trigger] --> Loader["Profile Loader<br/>registry.yaml → system"] Loader --> Standards["_shared + SaaS1/SaaS2<br/>review standards"] Standards --> Composer[Prompt Composer] Composer --> AI["Gemini 2.5 Pro<br/>(pluggable provider)"] AI --> Reflect["Self-reflection validation<br/>(optional)"] Reflect --> Comment[Post MR comment]
My Role
From v1.0 design through v2.0 expansion: profile module, the self-reflection pass, and the cross-microservice rollout.
Impact
- v1.0 → v2.0: 15 → 32 services
- Internal survey of 11: 82% read every AI comment, 73% changed code based on it
Lessons — A 9-year architecture lineage
2015 at iPanSec — A4P: Python subprocess running MobSF, web crawler scraping the report → structured output. 2024 at Cedars — AI Code Review: replace MobSF with Gemini API; the rest of the skeleton barely changed.
A senior engineer’s long-term value isn’t the new framework they can name. It’s recognizing which old problem just got a better solution.